Marketing Using Social Networks: Marketing Using Social Networks: Practice and Theory

Alan L. MontgomeryAssociate ProfessorAssociate Professor

The University of Hong Kong &Carnegie Mellon University

email: alanmontgomery@cmu.edu

All Rights Reserved, © 2012 Alan MontgomeryDo not distribute or reproduce without Alan Montgomery’s Permission

Lecture Outline

• PracticeMarketing Usage of Social Media– Marketing Usage of Social Media

– Measuring Social Media Influence

• Theory• Theory– Homophily or Influence? An empirical analysis of purchase

within a social network– Problem– Framework

Results– Results

2

Marketing UsageMarketing Usageof Social Media

Types of social media

• Discussion Forums– Amazon eBay Usenet Groups epinionsAmazon, eBay, Usenet Groups, epinions

• Blogs– Blogger, blog, Tumblr, Weebly, Xanga, LiveJournal, WordPress

• Wiki’s– Wikipedia, Knol

• Social Networks• Social Networks– SixDegrees.com (early effort in 1997, shutdown in 2000)– Open Social Networks: MySpace, Facebook, Bebo, Club Penguin– Vertical Network Sites: LinkedIn– Mobile social networks: Dodgeball

• Video- photo- and news-sharing sites• Video , photo , and news sharing sites– FlickR, YouTube, Digg 4

Marketing use of Social Media

• Forrester Research reports US online advertising spending in 2009 was approximately $25b or 12% ofspending in 2009 was approximately $25b or 12% of overall advertising budgets

• Most of the online budget went to search and display• Most of the online budget went to search and display advertising

• Social media is only $560m or less than 3% of the USSocial media is only $560m or less than 3% of the US online advertising budget in 2009

• Expected to almost triple to $1.76b by 2013p p $ y

5

BzzAgent: HTC Windows 7

• Goal: Increase Awareness, Drive recommendations, drive trial

• Strategy: Sample Experience1000 T-Mobile or AT&T customers (age 18-49 who were not current i-Phone

) i f HTC h tusers) were given a free HTC phone to try out

• Results: Reach, High-Volume Activity, Brand PerceptionBrand Perception

– Reached 234,000+ people via in-person and digitial conversations

– Each Agent influenced 235 conversations– Drove activity among core HTC segment:

25-34 year olds, influenced 499 Facebook posts and 651 Tweets

– Increased brand opinion by 36 Source: http://www.bzzagent.com/blog/post/htc/

p ypercentage points

– Increase advoacy levels by 85 percentage points

6

Coke and the 2008 Beijing Coke and the 2008 Beijing Olympics

• ChallengeAs the official sponsor of the Olympic Games, Coca-Cola needs to find an innovative way to amplify the brand message by connecting young targets to celebrate the first ever Olympics in China, and to bond the brand tightly to this special moment.

• ApproachApproachAt the night of March 24, 2008, Coca-Cola China launched a breakthrough digital activation on QQ I t t MQQ Instant Messenger (iCoke.qq.com, the IM platform with the largest user base in China), the first ever Coca-Cola Olympic Virtual To ch Rela ("VTR" fo sho t)Torch Relay ("VTR" for short).

7http://awards.wwwins.com/2009spikes/ico_votr.htm

OldSpice on YouTube and Twitter

• Old Spice body wash saleswash sales increased 107% in one month due o e o t dueto social media marketing (55% over past three months).

8

Ford Fiesta

• For used a campaign centered on user-generated content to ggenerate buzz and interest around the US launch of its Fiesta model.

• 100 individuals test-drove a Ford Fiesta for six months

• Document their experiences and ocu e t t e e pe e ces a dtravels in the car on social media networks

• Content was personalized andContent was personalized and authentic since it came from actual users

• Ford generated a 50 000• Ford generated a 50,000 member email list to further promote the car 9

Nike+

• Nike+ community allows users to track

• In Feb 2011, it had more than 2 million members who logged in allows users to track

their runs and challenge friends to distance and

ggover 336million miles.

• Sponsors the Human Race 10K• According to SportsOneSourcee ds to d sta ce a d

time running competitions.

According to SportsOneSource, Nike’s share of running shoes increased from 48% in 2006 to 61% in 20086 % 008

10

“United Breaks Guitars”

• Dave Carroll, who had been making a modest living as a musician

f i f ld i i h hiperforming pop-fold music with his band, Sons of Maxwell

• On March 31, 2008, Carroll and members of the band were flyingmembers of the band were flying from their hometown of Halifax, Canada, for a week of shows in Omaha, Nebraska

• During a connection in Chicago, other passengers aboard the flight noticed some very rough handling of cargo and Carroll’s bandmatesof cargo, and Carroll’s bandmateswatched helplessly as Dave’s $3,500 Taylor guitar was mishandled by United’s baggage handlers.

11

gg g

Social Networks

“This new organizational framework for online communities transformed the earlier topic drivencommunities transformed the earlier topic-driven models like forums to user-centric networks…with the individual at the center of their own community. This d dua at t e ce te o t e o co u ty smore accurately mirrors unmediated social structures, where ‘the world is composed of networks, not groups”, D.M. Boyd and N.B. Ellison (2007), Journal of Computer Mediated Communication

12

Reluctance by Some BrandsReluctance by Some Brandsto Use Social Media

• Cost and time– Companies fear that social media may require too many creative staff and p y q y

too much time

• Knowledge risk– Senior managers are less familiar with new media, so they avoid itg , y

• Incentive structure– Advertising firms with traditional media expertise have strong incentives to

maintain the current fee structurea a u u u

• Measurement– Gross rating points and click-through rates are more comfortable measures

than page views or engagementthan page views or engagement

• Loss of control– Marketers are accustomed to taking a top-down approach to brand positioning and

fear loss control of their brands

13Source: Sunil Gupta, Kristen Armstrong, and Zachary Clayton (2011)

Measuring Social Media Measuring Social Media Influence

Using Blogs

Message #1199 Civic vs

to Detect Market StructureAudi A6 Honda Civic 252Audi A6 Toyota Corolla 101Honda Civic Audi 6 252Honda Civic Toyota Corolla 2762

Message #1199 Civic vs. Corolla by mcmanus Jul 21, 2007 (4:05 pm)Yes DrFill, the Honda car model is sporty reliable and economical vs y

Toyota Corolla Audi A6 101Toyota Corolla Honda Civic 2762

sporty, reliable, and economical vs the Corolla that is just reliable and economical. Ironically its Toyota that is supplying 1.8L turbo ... Neon to his 16 year old brother I drove it A i ti N t k

Audi A6 Honda Civic Toyota

to his 16 year old brother. I drove it about 130 miles today. Boy does that put all this Civic vs. Corollaback in perspective! The Neon is

d l d i d d b ilt

Associative Network

Audi A6 Honda Civic yCorolla

Audi A6 --- 252 101

very crudely designed and built, with no low ...

Honda Civic 252 --- 2762

Toyota Corolla 101 2762 ---Corolla

Source: Netzer, Feldman, Fresko and Goldenberg (2011) “Mine Your Own Business”

The Car Models Network

Spring embedded Kamada Kawai graph g p

only ties with significant lifts  (2 p.value<0.01) are included

HokeyPokey.inHokeyPokey.inA Social Media Experience for an

Indian Ice Cream ParlorIndian Ice Cream Parlor• Goal

– How to use social media toHow to use social media to help this ice cream parlor to market itself?

• Method• Method– Quantify individual’s

influence in a social networkf– Predict impact of customer’s

influence– Measure net value of the

individual’s influence

• Implementation– “Creations on the Wall”Creations on the Wall– “Share your brownies” 17

Source: V. Kumar et al “Creating a Measurable Social Media Marketing Strategy for Hokey Pokey”18

Source: http://techtv.mit.edu/collections/isms:2635/videos/18090-hokey-pokey

Perils of Social Coupons

• Can substantially increase new customersnew customers

• However, the margin of these new customers is very low (Groupon charges 50%low (Groupon charges 50% of promotion cost)

• Can take over a year –perhaps even 8 years for beauty salon (if the customer is retained) tocustomer is retained) to recover from initial losses

S K d R j (2012) “S i l Source: Kumar and Rajan (2012) “Social Coupons as a Marketing Strategy”, Journal of the Academy of Marketing Science 19

Using Social Media to Using Social Media to Influence Activity

• Disney/ABC wants to know how social media reflects and influences consumers’ television viewing

• Goal is to develop metrics of psocial media to predict and influence audience viewership

• Ultimately wish to influence U t ate y s to ue ceviewership

20

E l f S i l M di F d Example of Social Media Feeds for “Revenge”

Source Subject

TwitterThis show is awesome! I hope it gets picked up for a full season. http://t.co/YjLz5EMG @GetGlue @RevengeN th t ll h fi i h d ll I h l ki f d t

Grey's Anatomy's Facebook Wall

Now that all my summer shows are finished, all I have looking forward to is The Vampire Diaries, Modern Family, Gossip Girl, Castle, Revenge and ofcourse Grey's Anatomy......only 7 more days for Grey's Anatomy ‐ Special Two‐Hour Season Premiere Event!!!! :))))

Twitter #Revenge could possibly be my new favorite showTheNest What Fall Shows Are Going to Make It?

T i@Gabriel_Mann @Revenge ahhh I can't wait! Already obsessed with the h !!Twitter show!!

Twitter @miklyn12 I'm loving this show so far! #Revenge

Twitter@smashleybell Can't wait to watch Revenge on my DVR! Hope to feature more of you on The Vogue Diaries! Love your style!Twitter more of you on The Vogue Diaries! Love your style!

Twitter@Revenge ahh already intense #RevengeABC a great start to the new fall shows:)Out of all of the shows I've started watching this fall @Revenge is my f b f d !Twitter favorite. Intrigue, subterfuge, and an amazing cast!

TwitterBut I have watched the first 20 minutes of Revenge, which is about as terrible and CWesque as one might expect.

21

Is this a causal relationship?

22Source: http://blog.nielsen.com/nielsenwire/online_mobile/the-relationship-between-social-media-buzz-and-tv-ratings/

Homophily or Influence?p yAn Empirical Analysis of

Purchase within a Social NetworkPurchase within a Social Network

Liye Ma, Alan Montgomery, and RamayyaKrishnan

23

Problem

Research Question

Why?Why?

25

Potential Reasons

Homophily Influence

Looks good Looks good Looks goodThis looks good

26

Why Do We Care?Promotion

Looks good Like this?Looks good

27

Why Do We Care?Promotion

Looks good This looks good

Refer a Friend?

28

Research Objective

• Quantify the effect of homophily and social influence in consumer’s purchase decisionin consumer s purchase decision.

29

It Can Be Tricky…

• Manski (1993)Correlated effects– Correlated effects

– Endogenous effects– Exogenous effectsExogenous effects

– Identification!

30

Data

What Data Do We Need?

• Consumers who are connected …• Communicate with one another…• And purchase products…p p• Over time…

• Identification: homophily is static, while i fl i d i ti tinfluence is dynamic, contingent on communication.

32

Data

• Obtained from a large Indian telecom companyOver 3 month period– Over 3-month period

– Cell-phone call records of more than 3.7 million users with 300 million phones calls in a major citiyj

• Observe:– Caller phone number– Callee phone number– Date and time of call

Length of the conversation– Length of the conversation– The purchase records of caller-ringback-tones (CRBT)

33

Product – CRBT

Caller ringback tones

34

Data – Preliminary AnalysisData – Preliminary AnalysisCan we do better?

B

Can we do better?

ACC

D

Adopter Non Adopter

35

Adopter Non-Adopter

Maybe, but needs modeling now

Data – Preliminary AnalysisData – Preliminary Analysis

Consumer Adoptions By Degree

1000000 0.7

10000

100000

onsu

mer

s

0.5

0.6

rcen

tage

100

1000

umbe

r of C

o

0.3

0.4

Adop

tion

Per

10

100

0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 90+

N

0.1

0.2 A

d d h l l ( d f l k)

Degree

Number of Consumers Adoption Percentage

36Node degree helps a lot (need for social network)!

Model

Exposure To Influence

If A calls B, and B has a certain CRBT, then we know A is exposed to it.

Purchase Incidence: Song-Change

Product Choice: Song

ktgipigiktgi

pigiktgi EEE ,,,1,,,

~)1( Smooth it out:

38smoothed exposure “raw” exposureparameter

Model

G groups of consumers

I consumers in each group

T time periods

J products

39

Consumer Decision –Consumer Decision –Purchase Incidence

Proportional hazard-rate, with time-varying coefficients

, ,( ) (1 ) exp( )gi gi t gi tS t t t

)exp( ,,,,,,, OutgiOutgiIngiIngigigi EE Out-group exposure

Intrinsic purchase rate IIntrinsic purchase rate In-group exposure“Susceptibility” to influence

40“when-to-buy”

Consumer Decision –Consumer Decision –Product Choice

Discrete-choice modelIn group exposureIn-group exposure

, , , , , , , , , , , ,gi j t j gi gi In gi j t In gi Out gi j t Out gi j tU E E X β, , , , , , , , , , , ,gi j t j gi gi In gi j t In gi Out gi j t Out gi j t

Intrinsic utility Out-group exposure“Susceptibility” to influence

Multinomial Logit

41“what-to-buy”

Homophily

In-group correlation of parameters

)1

(~ 21

rrrr

MVNg

)(~ 2 N )1

...,...(~...

rr

rrMVN

gI

),(~ Ni

0r Indicates homophily

Each of the parameters:

42

Hierarchical Bayesian Model

r2Population

Group/Consumer11g 21g Ig1

…Group/Consumer

12g 22g Ig2…

…

Data 1gD 2gD IgD… 1gD 2gD IgD……11g 21g Ig1… 12g 22g Ig2…

43

Model Summary

Two Components about consumer behaviorWh t h ?• When to purchase?– Proportional hazard model– Relates exposure to others within a group and outside of a– Relates exposure to others within a group and outside of a

group as well as baseline factors to predict when a purchase is made

• What to buy?– Multinomial choice model

R l t h t i ti ll i d t f– Relates song characteristics as well as in-group and out-of-group exposure to which song genre’s will be chosen

• Framed in a hierarchical Bayesian model to allow forFramed in a hierarchical Bayesian model to allow for individual variation 44

Results

45

Groups

1 3

5 79

6…

6 8…

5

2 68 9

3… 3 …

N-cliques

46

Findings

• Strong social influence effects in both purchase-timing and product-choice decisionsproduct choice decisions

• If we distinguish influence of in-group versus out-group we find that former is more important in purchase timing, while the latter is more important in product choice decisionlatter is more important in product choice decision– E.g., although customers are subject to influence by their friends,

this may be moderated by a desire not be perceived as imitating

• Strong homophily effect in the product-choice decision, where customers who are close by tend to have similar product tastes as well as similar susceptability to influencep y

47

Policy Simulation

Purchase Probability Enhancement

25 00%

30.00%

35.00%

40.00%

urch

ase

y

10.00%

15.00%

20.00%

25.00%

rove

men

t in

PPr

obab

ility Full Model

Homophily-OnlyInfluence-Only

0.00%

5.00%

10 20 30 40 50 60 70 80 90 100

C P tIm

pCoupon Percentage

4-10% Improvement In Purchase Probability

48

Policy Simulation

Choice Probability Enhancement

0 3y

0.2

0.25

0.3

ice

Prob

abili

ty

Full Model

0 05

0.1

0.15

vem

ent i

n Ch

o

Homophily-OnlyInfluence-Only

0

0.05

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Coupon Value

Impr

ov

12-35% Improvement In Product Choice Probability

49

Conclusions

50

Conclusions

• We are starting to see a shift in marketing emphasis on social media This is not new word of mouth hason social media. This is not new, word-of-mouth has always been a dominate way of promotion

• What is different is that marketers can now directly• What is different is that marketers can now directly measure it, participate in it on a massive scale, and influence it.

• Applications and research on social network influence is still early, but results show that it is impactful if used properly.

• It is critical to understand the difference between l ti d lit i thi t t B thcorrelation and causality in this context. Both are

present, but have very different implications. 51